The Use of Agent-Based Modeling for Evaluating Community-Level Obesity

Interventions

Implementation and Improvement Science webinar series co-hosted by the UCLA, USC and UCSD CTSAs and Kaiser Permanente

May 4, 2016

May C. Wang, MA, MPH, DrPH Professor, Department of Community Health Sciences

Roch Nianogo, MD, MPH PhD Candidate, Department of Epidemiology

Fielding School of Public Health University of California at Los Angeles

Outline

• Applying systems science – What is systems science?

– Why is it relevant (to obesity research)?

– What can systems science do?

– Agent-Based Modeling

• Application of ABM for obesity/intervention research: ECOSyS Study

What is Systems Science? • Systems science provides methods to address

complex systems. (Agent-based modeling is one such method)

• A complex system is “one whose properties are not fully explained by an understanding of its component parts” (Gallagher and Appeneller, 1999)

• “the whole is more than the sum of its parts because of the interactions between the parts” (Bonabeau E, PNAS, 2002)

Why is systems science relevant to obesity research?

Obesity: A Complex System • Multiple factors

• Dynamic interactions

• Feedback loops

http://www.shiftn.com/obesity/Full-Map.html

The Complexity of Obesity Research

Food Consumption

Physical Activity

Chronic stress

Zoning

Family Environment

Retail Food Environment

Childcare/Schools

Food System Food Assistance Programs

Health Care Providers

Some factors that play a role in obesity development

Food Marketing (and regulations)

Agricultural Policy

Transportation

Parks & Recreational

Facilities Neighborhood

Walkability Work Places

The Socio-ecological Framework

Food Behaviors Physical Activity Chronic stress

Family Environment Friends

Childcare/Schools Workplaces Health-care

Food system Agricultural policy

Food assistance programs Food marketing

Zoning

Walkability Parks & recreational

facilities Retail food environment

Transportation

Public Policy

Community and Neighborhood

Organizational

Interpersonal

Individual

Complex Systems have the following properties…

• Large number of heterogeneous elements

• Elements interact

• Produce an emergent effect which differs from the effects of the individual elements

• Effect persists over time and adapts as circumstances change (dynamic behavior)

Implications for research Traditional Statistical Modeling

(not adequate for studying complex systems)

Correlation-based

Linearity, Normality, Homogeneity, Independence

Inherently reductionist

Interactions between two variables at a time

Computational Modeling Allows for feedback loops and scale-free distributions

Interactions among individuals and their environment

Implications for research

Computational modeling …

.. does NOT replace statistical modeling

.. adds to existing analytic tools

What Systems Science Can Do

1. Considers the big picture and context of complex problems

2. Examines dynamic interrelationships of variables at multiple levels of analysis (e.g., from cells to society) simultaneously:

- causal feedback processes

- impact on the behavior of the system as a whole over time

Source: NIH Office of Behavioral and Social Sciences Research. http://obssr.od.nih.gov/scientific_areas/methodology/systems_science/

What Systems Science Can Do

3. Simulates the impact of policy decisions and therefore help decide on the most effective way to address a problem

4. Helps understand why programs and interventions do not have their intended effects

Source: NIH Office of Behavioral and Social Sciences Research. http://obssr.od.nih.gov/scientific_areas/methodology/systems_science/

Some Systems Science Approaches

• Network Analysis

• Systems Dynamics

• Discrete event simulation

• Dynamic microsimulation modeling

Agent-Based Modeling

Agent-Based Modeling

• Within the system, there are “agents” that behave according to their individual properties, environment

• Considers interaction of behaviors emergent properties of the overall system

• Based on rules and characteristics about agents and their behaviors

• Uses computer simulation

• Investigates complex systems from the ground up

Fast food

Child-care

Community

WIC clinic

Agent-Based Modeling: Core features

Ref: Luke & Stamatakis, 2012. Ann Rev Public Health 33:357-376

Heterogeneous Agents allowed to differ from one another

Spatial Agents located in a defined space

Bounded rationality Agents assumed to have imperfect knowledge

Interactive Agents interact locally with each other and with the environment

Dynamic Recursive models, can change non-linearly, with feedback

An Example

1) “Recreate” the present 2) Make projections

Current model projections

Reality

Our new model projections

Time

% obese

Fast-food

Child-care

Community

WIC clinic

INTERVENTION

Impact of multiple interventions in diverse settings

Computer modeling

Virtual population

No intervention

Intervention B

Intervention A

Evaluating Community Interventions

• Use of the “quasi-experiment”

• Challenges

– Dynamic situation

– Unable to “control” what a community does

– Uses statistical modeling:

High “precision” but not realistic and poor fit

Statistical models vs. Agent- based models

Ip, E. H., Rahmandad, H., Shoham, D. a, Hammond, R., Huang, T. T.-K., Wang, Y., & Mabry, P. L. (2013). Reconciling statistical and systems science approaches to public health. Health Education & Behavior : The Official Publication of the Society for Public Health Education, 40(1 Suppl), 123S–31S. doi:10.1177/1090198113493911 Levins, R. (1966) The strategy of of model building in population biology. American Scientist 54:421-431.

Statistical models • High precision and moderate generalizability • Low realism and fit

Agent-based models • Moderate generalizability, fit and precision • High realism

Statistical models not adequate for evaluating community interventions

The Early Childhood Obesity Systems Science Study

(ECOSyS)

(Sep 2013-Aug 2018)

Funded by the

Eunice Kennedy Shriver National Institute of Child Health and Human Development, NIH

PIs

Michael Prelip & May Wang

Project Coordinator

Amy Wang

Graduate Student Researchers

Roch Nianogo, Tabashir Nobari, Brenda Robles, Linghui Jiang

PHFE-WIC

LACDPH

First 5 LA

Kaiser Permanente

Samuels Center

UC Berkeley

U Washington

U

Saskatchewan

Collaborators

Los Angeles County: Multiple efforts to address

childhood obesity in the same community at multiple levels

since the mid-2000s.

Public Policy

Community and Neighborhood

Organizational

Interpersonal

Individual

Examples of Interventions

Since 2005, considerable effort to address childhood obesity in LA County and California. For example:

• State/Regional/Local:

– HEAC , BHC (The California Endowment)

– Community Benefits Program (Kaiser Permanente)

– School wellness programs (California Project LEAN)

– School food policies (state, local)

– First 5 LA’s programs to promote breastfeeding and healthy eating among preschool-aged children

– Community Transformation Grants (CDC)

• National:

– New WIC food package mandate (2009)

Impact on Early Childhood Obesity

Some questions:

What intervention strategies are the most effective?

Are there synergistic effects?

In what context is an intervention strategy most likely to be effective?

Study Aims

1. Identify obesity-related interventions and policies in LA County since 2003

2. Develop and validate community-level “intervention dose index”

3. Estimate obesity trends in preschool-aged WIC participants

4. Evaluate relationships between preschool-aged obesity trends and community-level intervention dose

• Use multilevel modeling, causal inference methods

5. Apply agent-based modeling to explore dynamic interactions, feedback mechanisms, and efficacy of intervention strategies

Data Sources (2002-2013)

Child Obesity Rates

Data Mining Project (PHFE WIC)

LA County Community Interventions &

Policies

LACDPH data

Key informant interviews

Index of community intervention dose

LA County Neighborhoods

U.S. Census

Food store environment (Dun & Bradstreet)

Local planning department

Local police department

Workgroup 1: Constructs & Domains

Workgroup 2: Review variables Data Collection

Development of Community Intervention Dose

Identify LA County Community Interventions & Policies (2003-2013)

Two workgroups

• WG1: Identify relevant domains & constructs

– Physical resources, social resources, capacity development, programs and policies, context

• WG2: Operationalize constructs & develop data collection instruments/protocols

Analysis

• Use multi-level modeling and causal inference methods to evaluate relationships between various intervention strategies and early childhood obesity risk

• Build ABMs to explore dynamic interactions and feedback mechanisms, and predict early childhood obesity rates under various scenarios involving different intervention strategies.

ABM: An Example

Some References Hennessy E, Ornstein JT, Economos CD, Herzog JB, Lynskey V, Coffield E, Hammond RA. Designing an Agent-Based Model for Childhood Obesity Interventions: A Case Study of ChildObesity180. Prev Chronic Dis. 2016 Jan 7;13:E04. Orr MG, Kaplan GA, Galea S. Neighbourhood food, physical activity, and educational environments and black/white disparities in obesity: a complex systems simulation analysis. J Epidemiol Community Health. 2016 Apr 15. Burke JG, Lich KH, Neal JW, Meissner HI, Yonas M, Mabry PL. Enhancing dissemination and implementation research using systems science methods. Int JBehav Med. 2015 Jun;22(3):283-91. Nianogo RA, Arah OA. Agent-based modeling of noncommunicable diseases: a systematic review. Am J Public Health. 2015 Mar;105(3):e20-31. [Review] Ip EH, Rahmandad H, Shoham DA, Hammond R, Huang TT, Wang Y, Mabry PL.Reconciling statistical and systems science approaches to public health. HealthEduc Behav. 2013 Oct;40(1 Suppl):123S-31S. Gortmaker SL, Swinburn BA, Levy D, Carter R, Mabry PL, Finegood DT, Huang T, Marsh T, Moodie ML. Changing the future of obesity: science, policy, and action. Lancet. 2011 Aug 27;378(9793):838-47 Mabry PL, Marcus SE, Clark PI, Leischow SJ, Méndez D. Systems science: a revolution in public health policy research. Am J Public Health. 2010 Jul;100(7):1161-3.

Acknowledgments The ECOSyS Study is funded by the Eunice Kennedy Shriver National Institute of Child Health and Human Development (RHD072296A).

We would also like to thank First 5 LA for providing funding to support the development and maintenance of the unique WIC database for LA County, and also the following individuals who have contributed significantly to the Data Mining Project:

• Armando Jimenez, First 5 LA, Director of Research and Evaluation

• Eloise Jenks, WIC Executive Director (recently retired)

• Kiran Saluja, WIC Executive Director (current)

• Shannon Whaley, WIC Research and Evaluation Director

• Mike Whaley, WIC IT Director

• Judy Gomez, WIC Deputy Director

• Maria Koleilat, WIC Research Analyst

• Lu Jiang, WIC Research Analyst

We would also like to thank the American Heart Association for support of the initial collaboration between UCLA and PHFE-WIC, and Tony Kuo, Brenda Robles, Margaret Shih, and Paul Simon of the LA County Department of Public Health for providing preliminary data that allowed for the development of the ECOSyS.

“Today’s knowledge about something is not necessarily the same tomorrow. Knowledge is changed to the extent that reality also moves and changes. Then theory also does the same.

It’s not something stabilized, immobilized.”

Horton & Freire, 1990

Thank you!